FabHK

A friend of mine works in fashion. Very briefly for background: there are runway models (who walk on live shows), and catalogue models (who work in studios presenting clothes against a neutral background in standard poses, more or less as living mannequins). The latter are being increasingly replaced by AI.

A friend of mine runs a business similar to this for systems migration, which is a fairly tough and tedious "short-put" job (you have to know both old and new system fairly well, and if all goes well, you get the same job done on the new system, while if something goes wrong, you're in a world of pain...):

The AI analyses sample data from old and new system, and then builds a prototype deterministic mapping pipeline. The experts can then inspect said pipeline and with their domain expertise catch edge cases/exceptions/uncovered cases, fix them and improve things, and then you use the vetted deterministic pipeline for migration. Yet, the initial AI draft saves enormous time.

Here a few sources:

https://cybernews.com/crypto/north-korean-crypto-hackers-thi...

https://www.nknews.org/2024/03/north-korea-relies-on-crypto-...

https://www.koreaherald.com/article/10599201

https://www.nbcnews.com/world/asia/north-korea-stolen-billio...

https://www.state.gov/releases/2026/01/the-democratic-people...

https://msmt.info/view/save/2025/10/22/26294780-c396-407d-bb...

That is a very naive take. First, the trust in crypto is extraordinarily expensive, as Budish's famous "The economic limits of Bitcoin" paper shows.

Second, the crypto bros running Bitcoin have reverted 20+ blocks on at least two occasions when it suited them to do so, and Ethereum has famously undone the DAO hack. Oh, and then of course you have all the bugs in smart contracts that can't be fixed. So the notion that "code is law" is laughable, and the notion that it's desirable is misguided, too.

Third, there's the oracle problem. Keeping track of bits in a ledger is a solved problem and has been a solved problem for a long time. The issue is correlating it with the real world. In your example, what if the student pays, then the teacher doesn't teach? How is the money flow reverted? If the student can just revert it, what happens if the student pays, the teacher teaches, and the student then reverts the payment? You need the legal system, and external adjudication, one way or the other.

Crypto adds nothing but idle speculation and the circumvention of (largely sensible) rules and regulations.

It was easier than that. There was a premium for Bitcoin in China and South Korea (the "kimchi premium") for quite some time due to capital controls. People in those countries would buy Bitcoin domestically (driving up the price there), send them to exchanges abroad, and sell them there, thus having taken their money out.

If you wanted to run the arbitrage, you'd buy Bitcoin cheap abroad, send them to China or Korea, sell them expensive, and now you're stuck with fiat in China or Korea which you can't take out, either. BitMEX founder Arthur Hayes was closing the loop in the early days by frequently crossing the border from China to Hong Kong with a backpack full of cash. SBF and Alameda claimed to have made their early money similarly by finding a way to close the loop on the kimchi premium.

Indeed. They have instead been solved by mobile-phone-based systems like Mpesa in Kenya and neighbouring countries.

Good. I also used to look at crypto with curiosity, but now think that

a) blockchain has no technological merit (given that permissioned systems already allowed for robust distributed consensus even in the presence of byzantine nodes back in the late 1990's).

b) the one innovation it brings (permissionlessness) is overwhelmingly negative (who really needs the "no gatekeeper" property?).

c) blockchain is good for nothing except crypto and crypto kittens (due to the oracle problem).

d) crypto is good for nothing except speculation and the circumvention of rules and regulations (which, like it or not, are mostly sensible).

Concretely for example on your point (it "let's some people keep their savings safe from dictatorial countries") - this would be trivial to offer for any bank using 1980's technology (offer an USD account just with username or password, no KYC), but it would also obviously be illegal and be shut down.

Making it harder to shut down - that's the one innovation crypto brings. But shutting it down is what we should be working towards.

The point of crypto is a) idle speculation and b) the circumvention of (mostly sensible) rules and regulations.

Estimates are that over the past few years, more than a third of the DPRK's foreign exchange income came from crypto. The legal exports were less than $0.5 bn, the rest is from more murky sources, including $1-2bn of crypto hacks in recent years.

1) The coins belonged to customers, but were in Binance wallets (thus under control of Binance) - that's how exchanges work. You send them your coins, then you can trade.

2) You can't freeze Bitcoin transactions (well, the crypto bros running it could if their majority wanted to, but they don't want to), but some of the stolen coins were stablecoins like USDT and USDC.

And fiat-backed stablecoins are ipso facto centralised (someone needs to 1) take receipt of fiat and issue coins against that, 2) custody that fiat and hold on to it reliably, and 3) return the fiat when given back the coins). You can then with the private key send the coins back to the issuer, but the issuer can refuse to give you the fiat.

However, many stable coin smart contracts are even more powerful: they give the issuer an authorized control account that can add a wallet address to a blacklist. The contract checks that list whenever someone tries to move the token; if the address is blocked, it rejects the transfer. Then the balance can’t be moved.

There’s a small difference: USDC blocks transfers both to and from a blacklisted address. USDT generally blocks the address from sending USDT, though it may still receive USDT. Tether’s contract also has a separate function that can destroy tokens held at a blacklisted address; that is distinct from simply freezing them. Note that when they destroy tokens, Tether keeps the cash - so it's a transfer from the criminals (or whoever held the tokens) to Tether. (Tether can then re-issue the coins and hand them back to law enforcement or the victim, but that's a separate step.)

It's well understood that crypto (scams, hacks) has contributed a third to half of the DPRK's foreign exchange income over the last few years.

Indeed.

However, it does not necessarily apply to algorithmic stable coins (Terra) or crypto-backed overcollateralised coins (DAI). That's why I said "fiat-backed".

here's an interesting article by DataFinnovation arguing that blacklisting crypto is doomed to fail.

https://medium.com/chainargos/losing-the-blacklisting-race-e...

Arguably any fiat-collateralised stablecoin is ipso facto centralised and trusted (someone needs to 1) issue the coins only against fiat collateral and then 2) hold onto that collateral reliably and 3) return it upon presentation of the coin), and at step 3 they can always say "no, we won't give it to you".

In particular it's the depth of the abstraction stack divided by transistor count. Wirth's Law: "Software is getting slower more rapidly than hardware is becoming faster."

https://en.wikipedia.org/wiki/Wirth's_law

In other words, even for Apple, Harlow's razor applies...

Per family member, of course.

Right, so 0.42% and 0.16% failure rate, respectively.

In the US alone, people make some 600m trips a day, so that "reliability" would result in about 2.5m to 1m failures a day.

Disagree. It's very simple - you have zero (off), half (parking), and fully on. It's naturally in that order, and the icons are irrelevant.

You're right, there's a factor 1/2 missing. Should've studied my integrals.

Right. So, it'll disengage and then, by virtue of this thing called gravity, crash on top of said other car or pedestrian. Much better.

> There were more Telsas than BYD in Hong Kong!

This is historical artefact: HK with its short distances is very suitable to EVs, and the government initially supported EVs with tax breaks. In other words, people bought a lot of Teslas when it was the only game in town.

For new registrations, BYD overtook Tesla back in 2025 with a yoy growth of 67.8% vs Tesla -3.8%, and in the latest available month (June) had more than 4x as many registrations as Tesla.

It's not exactly market-dominating growth justifying a 300+ P/E ratio.

https://www.thestandard.com.hk/news/article/319010/EVs-make-...

https://kr-asia.com/pulses/155906

https://hkcarrank.com/jan-2026-private-car-registration-data...

https://www.td.gov.hk/en/transport_in_hong_kong/transport_fi... -> check table 4.1(e)

The old pilot adage: a good landing is one you can walk away from. A very good landing is when you can re-use the plane.

Interesting. Air is approx 1 kg/m^3, so that tank is 0.4 kg at 1 bar, and a typical COPV can go to say 300 bar, so we're talking 120 kg.

For a 1000 kg car and with 200 m/s nozzle exit velocity, to hover the car you need (with m_0 g = m_a v/t) about 50 kg of air per second, so we're talking about roughly 2.5 seconds of flight (from those 400 litres).

There are three configurations:

- Performance: road ready, seats two, active aero and a “multi-mode suction system.” 0-60 mph in under 1.6 seconds. Available by reservation.

- Thrust: race ready, seats two, adds rear-mounted cold-gas thrusters. 0-60 mph in under 0.9 seconds. Limited release.

- Rocket: seats one, adds thrusters around the perimeter of the car for upward and rotational movement. 0-60 mph in under 0.59 seconds and a top speed of over 230 mph. Invite only.

Per Electrek: https://electrek.co/2026/10/15/tesla-unveils-cyberfied-new-r...

I wanted to learn Fortran in the 1990s and borrowed a Fortran book from the university library. On the first or second page it stated "To mark comments, just punch a letter 'C' into the first column of the punch card." I returned the book and never learned Fortran.

Having said that, so many foundational libraries were implemented in Fortran (BLAS, LINPACK/EISPACK, LAPACK, MINPACK, FISHPACK, FFTPACK, etc.) - thanks for that.

Sure, a given LLM can only produce so much (though it might be a lot), and then you'll need fresh inputs. But you contend that those have to be human, while I am arguing that they could be

- random,

- from the environment,

- from other AIs.

That increases the space considerably.

In particular, note that your argument also applies to humans: A human, being finite, can only produce finitely many things. But you add other humans, the environment, and pure chance to the mix, and it'll be pretty infinite for all practical purposes.

> Eventually it will run out of things that it can do with its training and the only way to get it to do more will be training it on new human found knowledge.

How do you know this?

The elderly can't handle a debit card?

It would be straightforward to establish the right to a free basic bank account (with transaction limits) for the homeless & (formerly) unbanked.

Indeed. This is not Dario Amodei, despite the caption...

https://www.youtube.com/watch?v=DlTNN0gvkLM

Very funny, btw. "Can we please stop anthropomorphising the LLMs? They hate that."